SCM

[#6041] ProbABEL's Cox module reports too many errors (beta may be infinite, setting beta and se to 'NaN')

View Trackers | Bugs | Download .csv | Monitor

Date:
2015-03-06 10:50
Priority:
4
State:
Closed
Submitted by:
Lennart Karssen (lckarssen)
Assigned to:
Lennart Karssen (lckarssen)
Resolution:
Fixed
Operating System:
All
Severity:
major
Hardware:
All
Version:
PA v0.4.4
Component:
ProbABEL
URL:
 
Summary:
ProbABEL's Cox module reports too many errors (beta may be infinite, setting beta and se to 'NaN')

Detailed description
Reported via e-mail by Anne G.:
------------------------------------------------------------------------------------------------------------------------------------------
When using a constant covariate (=0) to the model (see bug #1266), I get the error message
X matrix deemed to be singular, setting beta and se to 'NaN'
for all SNPs, sometimes with an additional message
beta may be infinite, setting beta and se to 'NaN'
nr of iterations > MAXITER (20): 20
Cox regression ran out of iterations and did not converge, setting beta and se to 'NaN'
Consequently, all parameter estimations (beta, SE, chi^2) are nan for all SNPs.

When adding a non-constant covariate to the model (e.g. sex), I do get estimates, but I still get
X matrix deemed to be singular, setting beta and se to 'NaN'
Beta may be infinite, setting beta and se to 'NaN'
for many SNPs.

In this last case I do get parameter estimates for the SNPs with error messages, but those are usually marked by huge confidence intervals or unrealistic beta's. This is mostly for low-MAF SNPs.
------------------------------------------------------------------------------------------------------------------------------------------

This seems to be too many errors. We should double check the error detection with the R code in the survival package.

Followup

Message
Date: 2015-03-17 13:55
Sender: Lennart Karssen

Fixed in r1931.

Attached Files:

Changes:

Field Old Value Date By
ResolutionAccepted As Bug2015-03-17 13:55lckarssen
close_dateNone2015-03-17 13:55lckarssen
status_idOpen2015-03-17 13:55lckarssen
Thanks to:
Vienna University of Economics and Business University of Wisconsin - Madison Powered By FusionForge